A Learning Model of Evoked EEG Signals Based on PCA and Semi-supervised SVM

Since in the recognition of Evoked electroencephalograph (EEG) signal (P300 signal), there are many problems such as a large amount of data to be collected, multiple acquisition channels, complex calculation, difficulty in obtaining label samples by traditional recognition model and long training time, a learning model based on principal component analysis (PCA) and semi-supervised support vector machine (SVM) was proposed. In this model, the EEG signal is preprocessed first, and then the PCA method is used to reduce the dimensionality of the collected brain channels to obtain a set of optimal channel subsets that have a good response to the P300 signal. On this basis, the semi-supervised classification model based on SVM is designed, and improves on the traditional discriminant analysis method, through a Euclidean distance between tag data and no labels to obtain the relationship of them, and, in turn, to optimize the objective function, to solve the projection vector, feature extraction, and then vote classification approach to the final recognition. Finally, experimental verification is carried out. The results show that the method can obtain 84.3% of the correct classification results under 40% of the label data. It has a good application prospect in the context of only a small amount of label data and reduces the time complexity. It has greater application value for online systems.